Camera rescaling applied twice in dataparser and dataset
Hi,
Is there a reason for camera rescaling being applied twice - both in dataparser and dataset classes? Is scale_factor and downscale_factor supposed to mean different things? Thanks!
https://github.com/nerfstudio-project/nerfstudio/blob/492fa0e67b238a1ca3158a37845aec313e5deb23/nerfstudio/data/dataparsers/nerfstudio_dataparser.py#L276
https://github.com/nerfstudio-project/nerfstudio/blob/492fa0e67b238a1ca3158a37845aec313e5deb23/nerfstudio/data/datasets/base_dataset.py#L55
Yea, agreed thats a little odd. The scale factors are defined in two different configs. The the dataparser one, it is used to determine which data to load from preprocess sizes (must be powers of 2). The dataset one lets you apply an arbitrary scaling (float) on the already loaded data.
https://github.com/nerfstudio-project/nerfstudio/blob/91291bc52c2501c8c93378382416edc2693647a8/nerfstudio/data/datamanagers/base_datamanager.py
and https://github.com/nerfstudio-project/nerfstudio/blob/91291bc52c2501c8c93378382416edc2693647a8/nerfstudio/data/dataparsers/nerfstudio_dataparser.py#L51-L52
Since InputDataset does a deepcopy of the dataparser's camera, wouldn't it be erroneously applying the scaling factor to the camera twice? https://github.com/nerfstudio-project/nerfstudio/blob/492fa0e67b238a1ca3158a37845aec313e5deb23/nerfstudio/data/datasets/base_dataset.py#L54
I encounter the same issue here. I think it looks weird to apply the scale twice. Is there an update on this thread? Thanks.